Change Detection in High-Resolution Remote Sensing Images Using Levene-Test and Fuzzy Evaluation
Change Detection in High-Resolution Remote Sensing Images Using Levene-Test and Fuzzy Evaluation
复制标题
使用 Levene 测试和模糊评估进行高分辨率遥感图像变化检测
DOI：
10.5194/isprs-archives-xlii-3-1695-2018
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
H. J. Liu
中科院分区：
文献类型：
--
作者：
G. H. Wang;H. Wang;W. Fan;Yunhai Liu;H. J. Liu
抽象的。针对高分辨率遥感影像空间结构复杂、纹理信息丰富的特点，提出了一种基于Levene检验和模糊评判的变化检测方法。该方法首先对两幅经过预处理的重叠图像进行分割，提取光谱、纹理等特征，得到图斑。然后，统计经过Levene-Test处理的所有图斑的变化信息，得到候选变化区域，通过IHS变换提取色调信息（H分量），结合纹理信息进行变化向量分析。最后通过迭代的方法确定阈值，计算候选变化区域的主题度，确定最终变化区域。对江苏省某地区多时相ZY-3高分辨率影像的实验结果表明：Levene-Test通过提取差异较大的图斑作为候选变化区域，降低了计算量，提高了变化检测的精度，对差异较大的不变区域表现出较好的容错能力。将色调纹理特征与模糊评价方法相结合，可以有效地减少遗漏和缺失，提高变化检测的精度。
Abstract. High-resolution remote sensing images possess complex spatial structure and rich texture information, according to these, this paper presents a new method of change detection based on Levene-Test and Fuzzy Evaluation. It first got map-spots by segmenting two overlapping images which had been pretreated, extracted features such as spectrum and texture. Then, changed information of all map-spots which had been treated by the Levene-Test were counted to obtain the candidate changed regions, hue information (H component) was extracted through the IHS Transform and conducted change vector analysis combined with the texture information. Eventually, the threshold was confirmed by an iteration method, the subject degrees of candidate changed regions were calculated, and final change regions were determined. In this paper experimental results on multi-temporal ZY-3 high-resolution images of some area in Jiangsu Province show that: Through extracting map-spots of larger difference as the candidate changed regions, Levene-Test decreases the computing load, improves the precision of change detection, and shows better fault-tolerant capacity for those unchanged regions which are of relatively large differences. The combination of Hue-texture features and fuzzy evaluation method can effectively decrease omissions and deficiencies, improve the precision of change detection.